1) directionality problem: you often don't know if variable A is causing variable B to change or if variable B is causing variable A to change, or if there is mutual interaction. E.g., self-esteem and GPA are strongly positively correlated, but you don't know if self-esteem causes you to get good marks or if getting good marks causes you to have high self-esteem, etc...
2) the third variable confound: a confound is a variable that is not controlled for, that may be affecting the relationship between two variables, e.g., IQ affects the relationship between both self-esteem and GPA. If you have a high IQ you're more likely to have high self-esteem and get good grades